Memory frequency vs latency

[sazrocks]

Ram kit (actually a 2 x 4GB and a 2 x 8GB) is in sig.

 

So recently I overclocked my ram to 3000MHz with 15-16-16-28 timings. Stock is 2133MHz with 14-14-14-35 timings. Is my overclock better or worse than stock? Should I try again but this time sacrifice frequency for better timings? Which applications benefit from better timings and which benefit from higher clocks?

[sazrocks]

1 minute ago, wrathoftheturkey said:

 Besides, no task is going to noticeably benefit from higher RAM speeds other than synthetic benchmarks, and even then barely.

What about gaming and compute intensive applications?

[sazrocks]

5 minutes ago, wrathoftheturkey said:

Gaming, not at all. Maybe if you're running Prime95, a little, but numbers for something like editing or even rendering should barely budge.

Not at all? What about this video:

Do you have any evidence that memory overclocking doesn't affect things like rendering?

[WereCat]

9 minutes ago, wrathoftheturkey said:

Gaming, not at all. Maybe if you're running Prime95, a little, but numbers for something like editing or even rendering should barely budge.

 

That is not true. Modern games benefit quite a bit from faster RAM even with higher latency (not just on Ryzen).

21 minutes ago, sazrocks said:

Ram kit (actually a 2 x 4GB and a 2 x 8GB) is in sig.

 

So recently I overclocked my ram to 3000MHz with 15-16-16-28 timings. Stock is 2133MHz with 14-14-14-35 timings. Is my overclock better or worse than stock? Should I try again but this time sacrifice frequency for better timings? Which applications benefit from better timings and which benefit from higher clocks?

First of all, have you made sure you are stable? Have you tested with Memtest86? There must be no errors, just because it boots doesn't mean it works properly.

 

Increasing speed is tied with increasing latency, you want to find the sweet spot.

2133MHz at CL14-14-14-35 is quite bad. If you are really stable at 3000MHz CL15-16-16-28 then you greatly increased your RAM performance and actually LOWERED your latency because the clock speed bump up is very significant and you increased CL by just 1.

If you went something like 3000MHz CL16-16-16-35 that would be a minor improvement.

 

EDIT:

 

I would recommend you to test stability with Memtest86 if you haven't done that.

Use Aida64 Memory and Cache benchmark to see your speed and latency improvements.

 

[Spuriae]

Frequency is much, much more important than latency. Your real latency numbers are lower at higher frequencies and your bandwidth only increases with frequency.

 

http://www.crucial.com/usa/en/memory-performance-speed-latency

 

Frequency increases lower your true latency a lot faster than directly lowering your latency.

14 minutes ago, wrathoftheturkey said:

Gaming, not at all. Maybe if you're running Prime95, a little, but numbers for something like editing or even rendering should barely budge.

Makes about as much difference gaming as the CPU; that is to say people worry way too much about it. Better clocked RAM is usually way cheaper than a better clocked CPU, though.

 

The problem with lots of memory reviews is they do their benchmarks with a GPU bottleneck, which just like for CPU reviews hides the real results. The trends are still present.

 

https://www.techspot.com/article/1171-ddr4-4000-mhz-performance/page3.html

http://www.gamersnexus.net/game-bench/2677-bf1-ram-benchmark-frequency-8gb-enough/page-2

http://www.guru3d.com/articles-pages/amd-ryzen-7-memory-and-tweaking-analysis-review,10.html

 

[DocSwag]

Your OC is much better. Bandwidth increased considerably and your latency actually went down. If you have 2133 MHz and 3000 MHz ram both at cl14, the 3000 MHz ram will have lower latency since timings are measured in number of cycles. With higher frequency ram, you have more cycles per second and so 14 cycles will go by much quicker on 3000 MHz ram resulting in lower latency.

[sazrocks]

@DocSwag @Nimrodor

Thank you!

 

@WereCat

I ran one pass (only had time for one, 24GB takes a loooong time) of memtest86 and it only had one error. I will test again overnight tonight because of that error. I also ran SuperPI 32M which completed with no errors.

 

I will also run a memory benchmark and see what improvements I got.

[sazrocks]

21 minutes ago, wrathoftheturkey said:

Mate... CPU bottlenecks can cost 30-40FPS. Do you have any benchmarks indicating that RAM speed is going to affect performance by more than 2 FPS in a demanding game?

That video I linked you?

[Spuriae]

23 minutes ago, wrathoftheturkey said:

Mate... CPU bottlenecks can cost 30-40FPS. Do you have any benchmarks indicating that RAM speed is going to affect performance by more than 2 FPS in a demanding game?

The three links at the bottom of my post are memory benchmarks.

 

The price delta in CPU's that causes a 30-40fps decrease is usually more than what people spend on memory as a whole. With more realistic choices like the difference between a 1600/6600k and 1700/7700k, boosting your OC'd memory clock by 1GHz for less than half the price of the CPU upgrade is a pretty enticing option for gaming. The differences between the 1600 and 1700 @1080p are about the same as the memory differences.

[sazrocks]

@WereCat

Ran 3 passes of memtest, no errors; I'm just going to say that one error was caused by a minor solar flare.

Here are the benchmark numbers:

Stock:

Spoiler

OC:

Spoiler

Dang I didn't realize read bandwidth would increase so much.

Also note that I have a 4.6GHz CPU overclock.

[WereCat]

2 hours ago, sazrocks said:

@WereCat

Ran 3 passes of memtest, no errors; I'm just going to say that one error was caused by a minor solar flare.

Here are the benchmark numbers:

Stock:

  Reveal hidden contents

OC:

  Reveal hidden contents

Dang I didn't realize read bandwidth would increase so much.

Also note that I have a 4.6GHz CPU overclock.

That is very nice increase in bandwidth and a lot better latency.

I am on DDR3 so going from 1600MHz CL9 to 2400MHz CL10 gives me almost 10,000MB/s more bandwidth as well.